alterlab-pathml
Pass
Audited by Gen Agent Trust Hub on Apr 29, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions require the installation of the 'pathml' library and its associated dependencies using the 'uv pip install' command.
- [EXTERNAL_DOWNLOADS]: The skill uses the 'SegmentMIFRemote' transform to send data to the DeepCell API at 'https://deepcell.org/api/predict' for remote cell segmentation tasks.
- [REMOTE_CODE_EXECUTION]: The skill demonstrates the use of 'torch.load' for deserializing model weights. This operation uses the Python pickle module and can lead to arbitrary code execution if the weights files are obtained from an untrusted source.
- [PROMPT_INJECTION]: The skill processes metadata and pixel data from external pathology slide formats (e.g., SVS, NDPI, DICOM), which presents an indirect prompt injection surface. 1. Ingestion points: references/image_loading.md (SlideData.from_slide) and references/data_management.md (SlideData.from_hdf5). 2. Boundary markers: Absent. 3. Capability inventory: SKILL.md (package installation), references/data_management.md (filesystem write), and references/multiparametric.md (network API access). 4. Sanitization: No sanitization is described for processing metadata strings extracted from external imaging files.
Audit Metadata